INDEX

Explanations

mentions of the word "wer" with varying levels of activation

oai_token-act-pair · gpt-3.5-turbo

the word "wer" in various forms, signaling a focus on variations of that term

oai_token-act-pair · gpt-4o-mini Triggered by @bot

New Auto-Interp

Top Features by Cosine Similarity

Comparing With GPT2-SMALL @ 0-res-jb

Configuration

jbloom/GPT2-Small-SAEs-Reformatted/blocks.0.hook_resid_pre

Prompts (Dashboard)

24,576 prompts, 128 tokens each

Dataset (Dashboard)

Skylion007/openwebtext

Features

24,576

Data Type

torch.float32

Hook Point

blocks.0.hook_resid_pre

Architecture

standard

Context Size

128

Dataset

Skylion007/openwebtext

Hook Point Layer

Activation Function

relu

Embeds

PlotsExplanationShow Test FieldDefault Test Text

IFrame

Link

Not in Any Lists

No Comments

Negative Logits

 dogs

-0.64

 ghosts

-0.64

 PTSD

-0.64

 accompanying

-0.64

 Jinping

-0.62

 makeup

-0.62

 parenting

-0.61

esp

-0.60

âĹı

-0.60

 paramedics

-0.60

POSITIVE LOGITS

wer

4.76

WER

1.65

wered

1.31

wark

1.18

Wer

1.14

ws

1.12

swer

1.11

wen

1.10

wed

1.04

Activations Density 0.011%

mentions of the word "wer" with varying levels of activation

the word "wer" in various forms, signaling a focus on variations of that term

No Comments

No Known Activations